Calculate ratio of counts in PostgreSQL-CodePudding

I have a table mydata in a Postgres 14 database, with the following relevant columns:

ftype: an enum having foo, bar and baz
status: another enum having pending, failed, success

I want the success rate of different types. Success rate is basically: the number of rows where the status is success divided by total number of rows for that ftype.

Currently, I'm doing the following:

SELECT
    COALESCE(
        COUNT(CASE WHEN ftype = 'foo' AND status = 'success' THEN 1 END) / 
        NULLIF(COUNT(CASE WHEN ftype = 'foo' THEN 1 END)::real, 0)
    ,0)::real AS foo_rate,

    COALESCE(
        COUNT(CASE WHEN ftype = 'bar' AND status = 'success' THEN 1 END) / 
        NULLIF(COUNT(CASE WHEN ftype = 'bar' THEN 1 END)::real, 0)
    ,0)::real AS bar_rate,

    COALESCE(
        COUNT(CASE WHEN ftype = 'baz' AND status = 'success' THEN 1 END) / 
        NULLIF(COUNT(CASE WHEN ftype = 'baz' THEN 1 END)::real, 0)
    ,0)::real AS baz_rate,
FROM mydata;

Is there a better/more performant way? How can I optimize it?
Would using PARTITION in the query help?

CodePudding user response：

Since you are computing rates for all types, a pivoted result would be simpler:

SELECT ftype, CASE WHEN ct_success = 0 THEN 0 ELSE ct_success/ct_total END AS rate
FROM  (
   SELECT ftype
        , count(*) AS ct_total
        , count(*) FILTER (WHERE status = 'success')::real AS ct_success
   FROM   mydata
   GROUP  BY 1
   ) sub;

Works for a selection of types, too. (The benefit is smaller, though.) Filter early:

SELECT ftype, CASE WHEN ct_success = 0 THEN 0 ELSE ct_success/ct_total END AS rate
FROM  (
   SELECT ftype
        , count(*) AS ct_total
        , count(*) FILTER (WHERE status = 'success')::real AS ct_success
   FROM   mydata
   WHERE  ftype = ANY ('{foo,bar,baz}'::my_enum_type[])  -- here!
   GROUP  BY 1
   ) sub;

About the WHERE clause:

IN vs ANY operator in PostgreSQL

You may want to round() and/or use numeric to begin with. See:

About the aggregate FILTER clause:

Aggregate columns with additional (distinct) filters

Whether table partitioning might help is hard to tell from the minimal information we have. There are useful guidelines in the manual chapter "Overview" for "Table Partitioning". My educated guess is that it wouldn't benefit you.